Fin whale vocalization classification and abundance estimation

Abstract

Several thousand fin whale vocalizations from multiple fin individuals were passively recorded by a high-resolution coherent hydrophone array system in the Gulf of Maine in Fall 2006. The recorded fin whale vocalizations have relatively short durations roughly 0.4 s and frequencies ranging from 15 to 40 Hz. Here we classify the fin whale vocalizations and apply the results to estimate the minimum number of vocalizing fin individuals detected by our hydrophone array. The horizontal azimuth or bearing of each fin whale vocalization is first determined by beamforming. Each beamformed fin whale vocalization spectrogram is next characterized by several features such as center frequency, upper and lower frequency limits, as well as amplitude-weighted mean frequency. The vocalizations are then classified using k-mean clustering into several distinct vocal types. The vocalization clustering result is then combined with the bearing-time trajectory information for a consecutive sequence of vocalizations to provide an estimate of the minimum number of vocalizing fin individuals detected.